Repository navigation
Thinking models aren't completing responses #848
Description
Describe the bug
Google Gemini 2.0 Flash Thinking Exp. 1219 is not completing the complete response, and thus, only give incomplete or partial responses. See below examples.
20250311-0556-14.3432897.mp4
Please complete the following information):
请补全以下内容
- OS: [e.g. Windows]
- Browser: [e.g. chrome]
- Extension Version: [e.g. v2.5.8]
Activity
norrybul commentedon Mar 25, 2025
Basing your sample, in most cases, it's more likely a bug on the server side of the Gemini model rather than ChatGPTBox. The request was sent properly by "CGB", and it only retrieved that much of response from the model. For the standard Gemini-2-Flash I'd tried using, responses were always complete and stops only if token limit is reached. A simple prompt like "continue" does the completion. Even the o3-mini reasoning models at ChatGPT playground have this bad habit too.
Maybe others can help you further on other possibilities if you can retrieve some logs between CGB and the Gemini server replicating this issue.